Note that the upstream llama.cpp project has now completely deprecated GGML in favor of GGUF [1]. How should the repository and user models adapt to this? [1] https://github.com/ggerganov/llama.cpp/pull/2398